fix(voice-call): pin response model sessions

CHANGELOG.md

@@ -66,6 +66,7 @@ Docs: https://docs.openclaw.ai
 
 - Discord/subagents: preserve thread-bound completion delivery by keeping the requester-agent announce path primary and falling back to direct thread sends only when the announce produces no visible output. (#71064) Thanks @DolencLuka.
 - Browser/tool: give Chrome MCP existing-session manage calls a longer default timeout, pass explicit tool timeouts through tab management, and recover stale selected-page MCP sessions instead of forcing a manual reset. Thanks @steipete.
+- Plugins/Voice Call: pin voice response sessions to `responseModel` before embedded agent runs, avoiding live-session model switch failures when the global default model differs. Fixes #60118. Thanks @xinbenlv.
 - Gateway/sessions: recover main-agent turns interrupted by a gateway restart from stale transcript-lock evidence, avoiding stuck `status: "running"` sessions without broad post-boot transcript scans. Fixes #70555. Thanks @bitloi.
 - Codex approvals: keep command approval responses within Codex app-server `availableDecisions`, including deny/cancel fallbacks for prompts that do not offer `decline`. (#71338) Thanks @Lucenx9.
 - Plugins/Google Meet: include live Chrome-node readiness in `googlemeet setup` and document the Parallels recovery checks, so stale node tokens or disconnected VM browsers are visible before an agent opens a meeting. Thanks @steipete.

extensions/voice-call/src/response-generator.test.ts

@@ -4,6 +4,8 @@ import type { CoreAgentDeps, CoreConfig } from "./core-bridge.js";
 import { generateVoiceResponse } from "./response-generator.js";
 
 function createAgentRuntime(payloads: Array<Record<string, unknown>>) {
+  const sessionStore: Record<string, { sessionId: string; updatedAt: number }> = {};
+  const saveSessionStore = vi.fn(async () => {});
   const runEmbeddedPiAgent = vi.fn(async () => ({
     payloads,
     meta: { durationMs: 12, aborted: false },
@@ -23,13 +25,13 @@ function createAgentRuntime(payloads: Array<Record<string, unknown>>) {
     runEmbeddedPiAgent,
     session: {
       resolveStorePath: () => "/tmp/openclaw/sessions.json",
-      loadSessionStore: () => ({}),
-      saveSessionStore: async () => {},
+      loadSessionStore: () => sessionStore,
+      saveSessionStore,
       resolveSessionFilePath: () => "/tmp/openclaw/sessions/session.jsonl",
     },
   } as unknown as CoreAgentDeps;
 
-  return { runtime, runEmbeddedPiAgent };
+  return { runtime, runEmbeddedPiAgent, saveSessionStore, sessionStore };
 }
 
 function requireEmbeddedAgentArgs(runEmbeddedPiAgent: ReturnType<typeof vi.fn>) {
@@ -126,4 +128,39 @@ describe("generateVoiceResponse", () => {
 
     expect(result.text).toBe("Absolutely. Tell me what you want to do next.");
   });
+
+  it("pins the voice session to responseModel before running the embedded agent", async () => {
+    const { runtime, runEmbeddedPiAgent, saveSessionStore, sessionStore } = createAgentRuntime([
+      { text: '{"spoken":"Pinned model works."}' },
+    ]);
+    const voiceConfig = VoiceCallConfigSchema.parse({
+      responseModel: "openai/gpt-4.1-nano",
+      responseTimeoutMs: 5000,
+    });
+
+    const result = await generateVoiceResponse({
+      voiceConfig,
+      coreConfig: {} as CoreConfig,
+      agentRuntime: runtime,
+      callId: "call-123",
+      from: "+15550001111",
+      transcript: [{ speaker: "user", text: "hello there" }],
+      userMessage: "hello there",
+    });
+
+    expect(result.text).toBe("Pinned model works.");
+    expect(sessionStore["voice:15550001111"]).toMatchObject({
+      providerOverride: "openai",
+      modelOverride: "gpt-4.1-nano",
+      modelOverrideSource: "auto",
+    });
+    expect(saveSessionStore).toHaveBeenCalledWith("/tmp/openclaw/sessions.json", sessionStore);
+    expect(runEmbeddedPiAgent).toHaveBeenCalledWith(
+      expect.objectContaining({
+        provider: "openai",
+        model: "gpt-4.1-nano",
+        sessionKey: "voice:15550001111",
+      }),
+    );
+  });
 });

extensions/voice-call/src/response-generator.ts

@@ -4,6 +4,7 @@
  */
 
 import crypto from "node:crypto";
+import { applyModelOverrideToSessionEntry } from "openclaw/plugin-sdk/config-runtime";
 import { normalizeLowercaseStringOrEmpty } from "openclaw/plugin-sdk/text-runtime";
 import type { SessionEntry } from "../api.js";
 import type { VoiceCallConfig } from "./config.js";
@@ -202,6 +203,7 @@ export async function generateVoiceResponse(
   const sessionStore = agentRuntime.session.loadSessionStore(storePath);
   const now = Date.now();
   let sessionEntry = sessionStore[sessionKey] as SessionEntry | undefined;
+  let sessionEntryUpdated = false;
 
   if (!sessionEntry) {
     sessionEntry = {
@@ -209,16 +211,29 @@ export async function generateVoiceResponse(
       updatedAt: now,
     };
     sessionStore[sessionKey] = sessionEntry;
-    await agentRuntime.session.saveSessionStore(storePath, sessionStore);
+    sessionEntryUpdated = true;
   }
 
   const sessionId = sessionEntry.sessionId;
-  const sessionFile = agentRuntime.session.resolveSessionFilePath(sessionId, sessionEntry, {
-    agentId,
-  });
 
   // Resolve model from config
   const { provider, model } = resolveVoiceResponseModel({ voiceConfig, agentRuntime });
+  if (voiceConfig.responseModel) {
+    sessionEntryUpdated =
+      applyModelOverrideToSessionEntry({
+        entry: sessionEntry,
+        selection: { provider, model },
+        selectionSource: "auto",
+      }).updated || sessionEntryUpdated;
+  }
+
+  if (sessionEntryUpdated) {
+    await agentRuntime.session.saveSessionStore(storePath, sessionStore);
+  }
+
+  const sessionFile = agentRuntime.session.resolveSessionFilePath(sessionId, sessionEntry, {
+    agentId,
+  });
 
   // Resolve thinking level
   const thinkLevel = agentRuntime.resolveThinkingDefault({ cfg, provider, model });